Non-hyperbolic Iterated Function Systems: attractors and stationary measures

Abstract

We consider iterated function systems $\mathrm{IFS}(T_1,\dots,T_k)$ consisting of continuous self maps of a compact metric space $X$. We introduce the subset $S_{\mathrm{t}}$ of {\emph{weakly hyperbolic sequences}} $ξ=ξ_0\ldotsξ_n \ldots \in Σ_k^+$ having the property that $\bigcap_n T_{ξ_{0}}\circ\cdots\circ T_{ξ_{n}}(X)$ is a point $\{π(ξ)\}$. The target set $π(S_{\mathrm{t}})$ plays a role similar to the semifractal introduced by Lasota-Myjak. Assuming that $S_{\mathrm{t}}\ne \emptyset$ (the only hyperbolic-like condition we assume) we prove that the IFS has at most one strict attractor and we state a sufficient condition guaranteeing that the strict attractor is the closure of the target set. Our approach applies to a large class of genuinely non-hyperbolic IFSs (e.g. with maps with expanding fixed points) and provides a necessary and sufficient condition for the existence of a globally attracting fixed point of the Barnsley-Hutchinson operator. We provide sufficient conditions under which the disjunctive chaos game yields the target set (even when it is not a strict attractor). We state a sufficient condition for the asymptotic stability of the Markov operator of a recurrent IFS. For IFSs defined on $[0,1]$ we give a simple condition for their asymptotic stability. In the particular case of IFSs with probabilities satisfying a "locally injectivity" condition, we prove that if the target set has at least two elements then the Markov operator is asymptotically stable and its stationary measure is supported in the closure of the target set.